AI Calorie Counter from Photo
Send a meal photo, get calories and macros in 3 seconds. No food database. No typing. Works in Telegram.
How It Works
No signup form. No food diary. No hunting through a 2-million-entry database for "medium banana, raw."
📸 Snap the plate
Top-down photo works best. Keep a utensil or hand in the frame — it helps the AI estimate portion size.
🤖 AI analyzes in 3 sec
Vision models identify food items, estimate volume, and calculate calories + macros for each component.
📊 Track across the day
Your daily total updates automatically. At the end of the day you see calories, protein, carbs, and fat without ever opening a spreadsheet.
Who This Is For
🚫 People who hate manual logging
If you've quit MyFitnessPal or Lose It twice in 6 months, the problem was logging friction. This removes it.
🍽️ Eat out often
Restaurants rarely publish nutrition facts. Photo estimate works on any cuisine without chasing menu data.
📉 Weight-loss beginners
You want calorie awareness, but you don't want to weigh every ingredient. Good-enough consistent tracking beats perfect tracking you quit.
🥩 Macro-curious lifters
Get rough protein/carb/fat splits per meal without typing ingredient names. Pair it with the macro calculator for targets.
⏱ Busy schedules
Tracking has to cost less than 10 seconds per meal, otherwise it doesn't survive a week with kids, commutes, and Slack.
🧠 Habit builders
If the goal is awareness (not a spreadsheet), a rough real-time estimate for every meal is more useful than a perfect log for half of them.
Who This Is Not For
- Competitive physique athletes in the last 6 weeks of a contest cut. Food scale + weighed ingredients is the tool for that window.
- Clinical dietary protocols — renal diets, controlled carbohydrate plans for diabetes management, elimination diets requiring precise gram counts. Use a registered dietitian and a scale.
- Supplement tracking — vitamins, electrolytes, creatine dosage. Photo-based AI is for food, not capsules.
Photo-Based vs Manual Logging
| Photo-based (Nouri) | Manual logging (MFP, FatSecret, Cronometer) | |
|---|---|---|
| Time per meal | ~5 seconds | 2–5 minutes |
| Per-meal accuracy | ±10–20% | ±5% (with a scale) |
| Long-term adherence | High — low friction | Low — 74% quit within 30 days |
| Restaurant coverage | Native — works on any dish | Limited — depends on database entry |
| Mental overhead | Almost none | Constant — search, select, weigh |
| Barcode scanning | Not needed | Required for packaged food |
The right way to read this: per-meal, manual logging is more accurate. Across 90 days, photo-based tracking wins because it's still happening. Read our piece on why people quit calorie tracking apps for the data behind the adherence gap.
Limitations We're Honest About
- Dense mixed meals (soups, stews, curries) are hardest — the AI can't see below the surface, so it uses typical recipe ratios as a fallback.
- Liquid calories in restaurants (sauces, dressings, cooking oil) are often under-estimated because they're invisible. For restaurant meals, expect to add ~10% for cooking fat.
- Portion sizing without a reference object is less precise. Include a hand, fork, or drink glass when you can.
- Close-up shots of a single food item with no plate context can trip up volume estimation. Top-down full-plate shots produce the best estimates.
- Estimates, not lab measurements. Treat them as trend data. If the weekly average matches your target calories and weight moves predictably, the estimates are doing their job.
What You Actually See
A typical workflow inside Telegram — no screenshots needed to imagine it:
- You: [photo of oatmeal with berries and almond butter]
- Nouri: "Oatmeal + berries + almond butter. ~340 kcal · P 11g · C 48g · F 12g. Great breakfast — solid fiber. Want to log it?"
- You: "Actually closer to a cup and a half of oats"
- Nouri: "Updated: ~420 kcal · P 14g · C 60g · F 13g. Logged. Daily total: 420/2100."
That's it. The macro tracker side produces the same flow for protein/carb/fat targets. Combined, you get calorie awareness + macro tracking without ever opening a food database.
Frequently Asked Questions
How accurate is an AI calorie counter from a photo?
Nouri's photo-based estimates typically land within 10–20% of a careful manual logging result for everyday home-cooked meals. Accuracy is highest for single-ingredient items (grilled chicken, an apple, rice with obvious volume) and lowest for dense mixed stews or restaurant dishes with hidden oils and sauces. For most weight-management goals, that error is smaller than the natural day-to-day variation in how much you actually eat.
Can it count calories from mixed meals?
Yes. Nouri identifies distinct food items on the plate, estimates portions from visible volume, and sums macros across them. Complex stews or layered dishes are harder because the AI can't see ingredients buried inside, so it errs on the side of typical recipe ratios. For mixed meals, the estimate is practical but less precise than for meals where everything is visible.
Does it work with restaurant food?
Yes — photo-based tracking is especially useful for restaurants because most menus don't provide nutrition facts. Nouri estimates calories from the visible plate regardless of where you're eating. See our guide at /restaurant-calorie-checker/ for how to frame the photo to get the best estimate.
Is it better than manual calorie logging?
Better for consistency, not necessarily for precision. Manual logging with a food scale is more accurate per meal but so high-friction that most people quit within 3 weeks. Photo-based tracking is slightly less precise per meal but sustainable long-term — and consistent "good enough" tracking beats perfect tracking that stops.
Do I need to install an app?
No. Nouri runs entirely inside Telegram. Send a photo to the bot, get an answer. If you prefer, the same analysis is available on the web version.
What's the catch — is it really free?
The free plan covers the core photo-to-calories workflow with basic daily tracking. Pro ($9.99/mo) adds weekly insight reports, custom macro targets, and higher analysis limits. There's no time-limited trial — free stays free.
How does AI estimate portion sizes from a 2D photo?
Two signals: food items are identified by vision models trained on millions of labeled food photos, then portion size is estimated from context — plate diameter, utensils, drink glasses. Results improve when a reference object (hand, fork, coin) is in the frame.
Can I correct an estimate if it's wrong?
Yes. Reply to the bot with "that was closer to 300 calories" or "more rice than that" and Nouri updates the entry. The correction also helps Nouri calibrate its estimates on your specific cooking style over time.
Related Reading
Try It on Your Next Meal
Free. No credit card. 3 seconds to your first estimate.
Start Free in Telegram